a material world - simberian.com · a material world . modeling dielectrics ... t ∂ ∇× = + ∂...
Post on 17-Apr-2018
233 Views
Preview:
TRANSCRIPT
TITLE
Image
A MATERIAL WORLD Modeling dielectrics and conductors
for interconnects operating at 10-50 Gbps C. Nwachukwu, (Isola)
Y. Shlepnev, (Simberian) S. McMorrow, (Teraspeed-Samtec)
Outline • Broadband dielectric and
conductor models • PCB materials and model
identification techniques • Practical examples of material
model identification
Yuriy Shlepnev Simberian Inc. shlepnev@simberian.com
Chudy Nwachukwu Isola Group chudy.nwachukwu@isola-group.com
Scott McMorrow Teraspeed - Samtec scott@teraspeed.com
“Material world” terminology • Materials:
– Dielectrics… are the broad expanse of nonmetals considered from the standpoint of their interactions with electric, magnetic or electromagnetic fields. - A. R. von Hippel, “Dielectric materials and applications”
– Conductors are materials that allow the flow of electrical current
• Linear material satisfy superposition property: • Time Invariant material does not change behavior with time: • Material is passive if energy is absorbed for all possible values of fields for all time
• Material is homogeneous if properties do not change through some area/volume • Material is isotropic if properties do not change with direction • Material is anisotropic if properties change with direction • Temporal dispersion is momentary delay or lag in properties of a material usually observed as
frequency dependency of the material properties
( ) ( ) ( ) ( )x t w t x t w tτ τ→ ⇒ − → −1 1 2 2 1 2 1 2;x w x w x x w wα β α β→ → ⇒ ⋅ + ⋅ → ⋅ + ⋅
( ) ( ) ( ) 0,t
S
P t E H ds d tτ τ τ−∞
= × ⋅ ≥ ∀
∫ ∫
3
Maxwell’s equations in macroscopic form
4
freeDH Jt
∂∇× = +
∂
freeD ρ∇ ⋅ =
0B∇⋅ =BEt
∂∇× =−
∂
Gauss’s laws
Ampere’s law
Faraday’s law
E - Electric Field (V/m) H - Magnetic Field (A/m) D - Electric Flux (Coulomb/m^2)
B - Magnetic Flux (Tesla or Weber/m^2)
freeρ - Free Charge (Coulomb/m^3)
freeJ - Free Current (A/m^2)
0D E Pε= +
( )0B H Mµ= +
No material equations here…
P - Polarization (Coulomb/m^2)
M - Magnetization (A/m) Fields in materials
Currents in Ampere’s law:
DH Et
σ∂∇× = +
∂
V
-
+ E electric field
freeJ current
( ),T,...freeJ f E=
freeJ Eσ=
σ - bulk conductivity, Siemens/m
Translational motion of free charges in electric field:
Ohm’s Law for LTI, isotropic:
5
freeDH Jt
∂∇× = +
∂
dispersive in general; almost constant up to THz;
Conductivity current [A/m^2]
1 /ρ σ= - bulk resistivity, Ohm*m
Currents in Ampere’s law:
+ + + + + + + + + +
- - - - - - - - - - + + + + + - - - - - V
0D E Pε= +
PE
diaE
( ), H,T, F,...P f E=
electric field in vacuum polarization
smaller electric field in dielectric
0E PH Et t
ε σ∂ ∂∇× = + +
∂ ∂
average of dipole moments [Coulomb/m^2]
0 *P Eε χ=
-
+
χ - dielectric susceptibility (always dispersive)
for LTI, Isotropic:
A. R. Von Hippel, “Dielectrics and Waves”, 1954 B.K.P. Scaife, “Principles of dielectrics”, 1998
6
Polarization [Coulomb/m^2] is displacement of charges bound to atoms, molecules, lattices, boundaries,… - creates electric field
DH Et
σ∂∇× = +
∂
0limV
qdP
V→= ∑
Polarization Current – movements of bound charges
1 V, 50 Ohm
7-mil wide microstrip line on 4 mil dielectric (Dk=4.2, LT=0.02 @ 1 GHz); Segment 60 mil long in 1 mil thick layer Signal1; Instantaneous values at 1 GHz, t=0 computed with Simbeor THz
[V/m]
Electric Field Strength
Polarization Density
Polarization Current Density [A/m^2]
[C/m^2] 50 Ohm
Why electric field is larger in dielectric?
7
Polarization current is real current!
Dielectrics and Conductors • Dielectric temporal
dispersion – Debye model – Modifications of Debye
model – Multipole Debye model – Wideband Debye model – Lorentzian model – From DC to infinity
• Inhomogeneous dielectrics • Anisotropic dielectrics
• Conductor temporal dispersion – Skin effect – Conductor roughness
• Effective roughness layer • Modified Hammerstad model • Huray’s snowball model
– Advanced conductor models • Ferromagnetics • Breaking the skin…
8
Dielectrics vs. Conductors Dielectrics
• Electric polarization dominates
• Small number of free charges ~10^10 to ~10^16 1/m^3
• Small bulk conductivity ~10^-9 to ~10^-16 1/Ohm*m (large resistivity)
• Conductivity increases with the temperature
Conductors • Almost no electric polarization up to
~10^16 Hz (shielding) • Large number of free charges ~10^27
to ~10^29 1/m^3 • Large bulk conductivity
~10^6 to ~10^8 1/Ohm*m (small resistivity)
• Conductivity decreases with the temperature
Semi-m
etals Sem
iconductors
9 C.A. Balanis, Advanced engineering electromagnetics, 2012 I. S Rez, Y.M. Poplavko, Dielectrics (in Russian), 1989
Debye temporal dispersion ( ) ( ) ( )0
1P tP t E t
tεε
τ τ∂ ∆
+ =∂
Normalized impulse response (susceptibility kernel):
( ) , 0t
t e tτδ
εχτ
−∆= ≥
V ( )E t V R
C
Generalization - polarization for any excitation (convolution):
( ) ( ) ( )' ' '0
t
P t t t E t dtδε χ−∞
= − ⋅ ⋅∫
( ) ( )1 , 0t
h t e tτχ ε −= ∆ − ≥
Normalized step response: Normalized susceptibility for relaxation time 0.16 ns or frequency 1 GHz
( ) ( ) ( )1 1Q tQ t V t
t RC R∂
+ =∂
- relaxation time - difference between susceptibility at 0 and infinity
τε∆
( ) 1 tC t e
Rτ
δ−=
( ) ( )1t
hC t C e τ−= −
RCτ = - relaxation time
( ) ( ) ( )' ' 't
Q t C t t V t dtδ−∞
= − ⋅ ⋅∫
(effective capacitance)
P. Debye, “Polar molecules”, 1929. or H Frohlich, “Theory of dielectrics”, 1949.
10
Debye temporal dispersion in frequency domain
( ) ( ) ( )01i P P Eεω ω ω ε ωτ τ
∆⋅ + =
Normalized impulse response (susceptibility):
( )1 i
εχ ωωτ
∆=
+
V ( )E ω V R
C
Generalization - solution for any excitation in frequency domain (LTI, isotropic):
( ) ( ) ( )0P Eω ε χ ω ω= ⋅
( ) ( ) ( )1 1i Q Q VRC R
ω ω ω ω⋅ + =
( )1
CCi RC
ωω
=+
Effective capacitance:
( ) 0, i tF t F e ωω = ⋅
( ) ( ) ( )Q C Vω ω ω= ⋅
( ) ( ) ( )01P t
P t E tt
εετ τ
∂ ∆+ =
∂( ) ( ) ( )1 1Q t
Q t V tt RC R
∂+ =
∂
11
Generalization – Ampere’s law in frequency domain
( ) ( ) ( ) ( ) ( )0 0H i E i E Eω ωε ω ωε χ ω ω σ ω∇× = + +
( ) ( ) ( )00
1H i Eiσω ωε χ ω ωωε
∇× = + +
( ) ( )1rε ω χ ω= +
( ) ( )0
1rc iσε ω χ ωωε
= + +
- relative permittivity
- relative “complex” permittivity
0E PH Et t
ε σ∂ ∂∇× = + +
∂ ∂
120 8.8541878176 10ε −≅ ⋅ - permittivity of vacuum
(constant), by definition
( ) 0, i tF t F e ωω = ⋅( ) ( ) ( )0P Eω ε χ ω ω= ⋅
Not constant for all materials!!!
12
Permittivity of Debye dielectric ( ) ( )1 1
1r iεε ω χ ωωτ
∆= + = +
+( ) 1
1rr
fi f fεε ∆
= ++ ( )
1rr
fi f fεε ε∞
∆= +
+
ε∆ ε∞
0.5 ε⋅ ∆
Frequency, Hz
12rf πτ
=
( )Re rε
( )Im rε−
( )( )
Imtan
Rer
r
εδ
ε= −
Relative permittivity for relaxation time 0.16 ns or frequency 1 GHz
4.0; 0.2; 1rf GHzε ε∞ = ∆ = =
0.50.5
εε ε∞
⋅ ∆+ ⋅∆
Example:
Time, s
V
13
Plane wave in Debye dielectric ( )
1rr
fi f fεε ε∞
∆= +
+Example:
( ) ( ) 0 02 rf i f fπ ε ε µΓ = ⋅ ⋅ - plane wave propagation constant
Attenuation Np/m
Frequency, Hz
Phase delay s/m
Instantaneous electric field along the wave propagation (normalized)
1 GHz
10 GHz
4.0; 0.2; 1rf GHzε ε∞ = ∆ = =
E
H
14
Length, m
Empirical modifications of Debye model ( )
( )( )1r
iβα
εε ω εωτ
∞∆
= ++
Havriliak, S.; Negami, S. "A complex plane representation of dielectric and mechanical relaxation processes in some polymers". Polymer N8: p 161–210,1967.
( )( )1r i αεε ω εωτ∞∆
= ++
K.S. Cole, R.H. Cole, (1941) ( )( )1r i β
εε ω εωτ∞∆
= ++
Cole-Davidson relaxation
( )Re rε
tanδ
( )Re rε
tanδ
1α =1.5α =
0.5α =
Frequency, Hz
1α =
1.5α =
0.5α =
1β =1.5β =
1β =
1β =1.5β =
0.5β =
1α =4.0; 0.2; 1rf GHzε ε∞ = ∆ = =
Frequency, Hz
0.5β =
15
Cole-Cole plots
1α =
1.5α =
0.5α =
1β =
1.5β =
0.5β =
4.0; 0.2; 1rf GHzε ε∞ = ∆ = =
Cole-Cole model Cole-Davidson model
ε∞ (0)ε ε∞ (0)ε
(Debye) (Debye)
16
Multipole Debye model ( )
1rr
fi f fεε ε∞
∆= +
+ ( )1 1
Kk
rk rk
fi f fεε ε∞
=
∆= +
+∑ K relaxation poles, 2K+1 parameters
kε∆∑ ε∞
Frequency, Hz
( )Re rε
( )Im rε−
tanδ
Relative permittivity for relaxation frequencies 0.1, 1, 10, 100 GHz
Poles:
Time, s
4-pole example:
1 2 3 4
4.0; 0.05;0.1; 1; 10; 100; [GHz]
k
r r r rf f f fε ε∞ = ∆ =
= = = =
17
Plane wave in multipole Debye dielectric ( ) ( ) 0 02 rf i f fπ ε ε µΓ = ⋅ ⋅ - plane wave propagation constant
Attenuation Np/m
Frequency, Hz
Phase delay s/m
( )21 lS eω −Γ⋅=
Generalized transmission parameter for distance l:
( )1 1
Kk
rk rk
fi f fεε ε∞
=
∆= +
+∑
1 2 3 4
4.0; 0.05;0.1; 1; 10; 100; [GHz]
k
r r r rf f f fε ε∞ = ∆ =
= = = =
4-pole example:
Insertion Loss dB/m
18
E
H~f
Frequency, Hz
Can we just fit Dk & LT points with multipole Debye model? • 2 problems
The result is very sensitive to measurement errors (requires data points consistent with the model) Bandwidth is restricted by the first and the last frequency point
From Isola’s FR408HR specifications
( )Re rε
tanδMeasured
Fitted No data to build model above 10 GHz!
Strip line, ideal conductor 19 Computed with
Simbeor THz
Wideband Debye model ( )
1 1
Kk
rk rk
fi f fεε ε∞
=
∆= +
+∑ Continuous relaxation poles from 10^m1 to 10^m2
ε∞
Frequency, Hz
( )Re rε
( )tanδ ω
POLES
Independently derived in 2 papers: C. Svensson, G.E. Dermer, Time domain modeling of lossy interconnects, IEEE Trans. on Advanced Packaging, May 2001, N2, Vol. 24, pp.191-196. Djordjevic, R.M. Biljic, V.D. Likar-Smiljanic, T.K.Sarkar, IEEE Trans. on EMC, vol. 43, N4, 2001, p. 662-667.
2
12 1
10( ) ln( ) ln(10) 10
m
r miff
m m ifεε ε∞
∆ += + ⋅ − ⋅ +
ε∆
110m 210m
( )Im rε
Four parameters m1 and m2 are usually fixed to 4 and 12-13
, , 1, 2m mε ε∞ ∆
Example:
( ) ( )9 9
3.707; 1.108; 1 4; 2 13;
Re (10 ) 4.2; tan 10 0.02
m mε ε
ε δ∞ = ∆ = = =
= =
20
Plane wave in Wideband Debye dielectric ( ) ( ) 0 02 rf i f fπ ε ε µΓ = ⋅ ⋅ - plane wave
propagation constant
Attenuation Np/m
Example:
Phase delay, s/m
Frequency, Hz
2
12 1
10( ) ln( ) ln(10) 10
m
r miff
m m ifεε ε∞
∆ += + ⋅ − ⋅ +
( ) ( )9 9
3.707; 1.108; 1 4; 2 13;
Re (10 ) 4.2; tan 10 0.02
m mε ε
ε δ∞ = ∆ = = =
= =
Insertion Loss dB/m
( )21 lS eω −Γ⋅=
Generalized transmission parameter for distance l:
21
E
H
~f
Frequency, Hz
Wideband Debye model properties
4.2rε =
Frequency, Hz
( )Re rε
( )tanδ ω
POLES
2
12 1
10( ) ln( ) ln(10) 10
m
r miff
m m ifεε ε∞
∆ += + ⋅ − ⋅ +
110m 210m
m1 and m2 are usually fixed to 4 and 12-13
0tan :rand computed with and at fε ε ε δ∞ ∆
Example: 9
04.2; tan 0.02; 10 ; 1 4; 2 13;3.707; 1.108;
r f Hz m mε δε ε∞
= = = = == ∆ =
Dk and LT at one point is sufficient to define the model!
[ ]( )
2 1tan ln(10)Im
r m mL
δ εε
⋅ ⋅ ⋅ −∆ = −
( )[ ]( )
Re( ) 1 tan
Im lnrLL
ε ε δ
∞ = + ⋅
tan 0.02δ =90 10f Hz=
20
10
10L ln10
m
mifif
+= +
22
Cole-Cole plots 9
04.2; tan 0.02; 10 ; 1 4; 2 13;3.707; 1.108;
r f Hz m mε δε ε∞
= = = = == ∆ =1 2 3 4
4.0; 0.05;0.1; 1; 10; 100; [GHz]
k
r r r rf f f fε ε∞ = ∆ =
= = = =
Multi-pole Debye Wideband Debye
23
Definition of Wideband Debye with data from spreadsheet • Which point to chose to define the
model? • Ambiguous…
From Isola’s FR408HR specifications
( )Re rεtanδ
Measured
Fitted
Example: strip line, ideal conductor
Measured Fitted
24
Computed with Simbeor THz
Definition of Wideband Debye with data from spreadsheet
LT @ 5 GHz
Example: 1 inch of strip line, ideal conductor
Dk @ 0.1 GHz Insertion Loss / inch
LT @ 0.1 GHz
Dk @ 10 GHz Phase Delay / inch
25 Computed with Simbeor THz
Lorentzian temporal dispersion ( ) ( ) ( ) ( )
22 2
0 0 0 02 2P t P t
P t E tt t
δω ω ε ε ω∂ ∂
+ + = ⋅∆ ⋅∂ ∂
Normalized impulse response (susceptibility):
( ) ( )0 202
sin 1 , 01
tt e t tδωδ
εχ δ ωδ
−∆= − ≥
−
V ( )E tV R
C
( ) ( )0 202
11 sin 1 , 01
th t e t tδωχ ε δ ω ϕ
δ−
= ∆ − − + ≥ −
Normalized step response:
( ) ( ) ( ) ( )2
21 1Q t Q tR Q t V t
t L t LC L∂ ∂
+ + =∂ ∂
- damping factor (unit-less); - resonant frequency (radian); - difference between susceptibility at 0 and infinity 0ω
ε∆
L
Capacitor charge:
0 1 LCω =2R C
Lδ =
Cε∆ =
δ 21 1tan δϕ
δ− −
=
Normalized susceptibility or capacitor charge for resonant frequency 1 GHz
Time, s
0.25δ =0.5δ =
1.5δ =
ε∆
( )20
2 20 02i
ε ωχ ωω ω δω ω
∆ ⋅=
− +
I.F.Almog, M.S.Bradley, V.Bulovic, The Lorentz Oscillator and its Applications 26
Permittivity of Lorentzian dielectric ( ) ( )
20
2 20 0
1 12r i
ε ωε ω χ ωω ω δω ω
∆ ⋅= + = +
− + ( )2
2 2 2r
rr r
ff f i f f
εε ω εδ∞
∆ ⋅= +
− +
ε∆ ε∞
Frequency, Hz
0
2rfωπ
=
( )Re rε
tanδRelative permittivity for resonant frequency 1 GHz
4.0; 0.2; 1rf GHzε ε∞ = ∆ = =
2ε
δ ε∞
∆⋅ ⋅
Example:
Time, s
V R
0.25δ =
0.5δ =1.5δ =
0.25δ =0.5δ =
1.5δ =
0.25δ =0.5δ =
1.5δ =
27
V
( )E ω
Plane wave in Lorentzian dielectric ( ) ( ) 0 02 rf i f fπ ε ε µΓ = ⋅ ⋅ - plane wave
propagation constant
Attenuation Np/m
Example:
Phase delay, s/m
Frequency, Hz
Insertion Loss dB/m
( )2
2 2 2r
rr r
ff f i f f
εε ω εδ∞
∆ ⋅= +
− +
4.0; 0.2; 11.5 ;0.5 ;0.25 ;
rf GHzred curvesbluecurves
black curves
ε εδδδ
∞ = ∆ = == −= −= −
Generalized transmission parameter for distance l:
Absorption resonances
28
E
H
Generalized models of dielectric • Debye – Lorentzian without conductivity
• Generic rational model with complex poles (no conductivity)
( )2
2 21 1
( )21
N Kn k k
n k k k k
n
frf f fr i f fr fifr
ε εε εδ= =
∆ ∆ ⋅= ∞ + +
+ ⋅ ⋅ ⋅ −+∑ ∑
( )*
*1
1( )2
Nn n
n n n
R Rfs p s p
ε ε=
= ∞ + + − −
∑
2s i fπ= ⋅2n n np i fα π= + ⋅
n n nR Rr i Ri= + ⋅
From 2N+1 to 4N+1 variables to identify; Can be fitted to Dk and LT measured at N+1 - 2N+1 frequencies;
- complex frequency; - complex poles; - complex residues;
2N+3K+1 variables to identify Suitable for direct optimization
Both models enable easy frequency and time domain analysis! 29
Can we use specs to build generic rational model? • Better than Debye, but
The result is sensitive to measurement errors (requires dense data points) Bandwidth is still restricted by the first and the last frequency point
From Isola’s FR408HR specifications
( )Re rε
tanδMeasured
Fitted
No data to have reliable model above 10 GHz!
Strip line, ideal conductor, 5 points, 4 real poles 30
Computed with Simbeor THz
Dielectric models from DC to infinity
( )Re rε
Frequency, Hz
( )Im rε−
conduction relaxation resonances
…if one asks a fellow scientist [physicist] “what happens when EM radiation in the range from 10^-6 to 10^12 Hz is applied to those systems [solids]” the answer is usually tentative or incomplete… - G. Williams in F. Kremer, A. Schonhals, Broadband Dielectric Spectroscopy, 2003
( )2
2 21 10
( )21
N Kn k k
n k k k k
n
frf fi fr i f fr fifr
σ ε εε εωε δ= =
∆ ∆ ⋅= ∞ + + +
+ ⋅ ⋅ ⋅ −+∑ ∑
31
Polarization mechanisms
( )Re rε
Frequency, Hz
Atomic or ionic polarization of molecules (induced dipoles)
Electronic polarization of atoms or ions (induced dipoles)
Orientation/Distortion molecular polarization (permanent dipoles)
Macro-dipoles (charges on boundaries or in lattice)
D.D. Pollock, Physical properties of materials for engineers, 1982, v III C.A. Balanis, Advanced engineering electromagnetics, 2012
Increase of relaxation time
32
+ -
+ -
+ -
Dielectric constant at “infinity” • It is UNIT for all materials • Practically, we can use value at the
highest frequency of interest
( )Re rε
3.5ε∞ =
1ε∞ =
3.2ε∞ =2.8ε∞ =
2.3ε∞ =
Value at “DC” should be define to have accurate value at the lowest frequency of interest
33
Frequency, Hz
Causality
34
• Condition for the impulse response of susceptibility leads to Hilbert transform or Kramers-Kronig relations between the real and imaginary parts of the frequency-domain permittivity:
• Realness or impulse response: real part is even and imaginary is odd function of frequency
( ) 0 0t at tδχ = <
Kramers, H.A., Nature, v 117, 1926 p. 775.. Kronig, R. de L., J. Opt. Soc. Am. N12, 1926, p 547.
( ) ( ) ( ) ( )r iiε ω ε ω ε ω ε χ ω∞= + = + 0limPVω ε
εω ε
− +∞
→−∞ +
= +
∫ ∫
( ) ( ) ( ) ( )' '' '
' '1 1,i r
r iPV d PV dε ω ε ω ε
ε ω ε ω ε ω ωπ ω ω π ω ω
∞ ∞∞
∞−∞ −∞
−= + ⋅ = − ⋅
− −∫ ∫
( ) ( ) ( )( )
,1, 01, 0
t sign t ttsign t t
δ δχ χ= ⋅
− <= >
( ) ( ){ } ( ){ } ( ){ }
( ){ } ( ) ( )' ''
12
2 1
F t F sign t F t
F sign t PV di i
δ δχ ω χ χ
πχ ω
χ ω ωω π ω ω
∞
−∞
= = ∗
= → = ⋅−∫
Derivation:
Use of K-K equations to restore real part
35
Linear growth of loss over some band -> constant imaginary part of permittivity
( ) 8 110.02; 10 10i f from to Hzε =
Singularities ( )Re rε
Add Debye slopes -> Wideband Debye model!
Frequency, Hz Frequency, Hz
Another way to estimate causality
36
Front delay of the impulse response: frontL
Tcε∞=
4.2; tan 0.02; 1 ; 3.71;r rf GHzε δ ε∞= = = =Wideband Debye model:
Wideband Debye Front Delay
“Flat” model: 4.2; tan 0.02;rε δ= =
“Flat” Model Front Delay Violation of Causality in “Flat” model and 6 ps/inch delay difference!!!
1 inch strip line, no conductor and reflection losses; 1 ps rise and fall, +2.74 ps delay;
See more at: M. Tsiklauri et al., Causality and Delay and Physics in Real Systems, IEEE Int. Symp. On EMC, 2014, p. 962-966.
There must be no response before the Front Delay!
or min phase delay for S-par.
Time, ns
Computed with Simbeor THz
Inhomogeneous dielectrics • Practically all PCB/packaging materials are heterogeneous mixtures of components • Two ways to deal with material the inhomogeneity:
– Direct electromagnetic analysis – specify separate material models for homogeneous regions (too many parameters – not practical);
– Homogenization – build macroscopic models for regions with fewer parameters; • Two ways to build macroscopic models:
– Empirical way – fit a broadband homogeneous model to measured data (easy); – Use mixing formulas or algorithms: construct macroscopic model from models of components if
component models and mixture parameters are known:
37
- broadband model of mixture
( )1ε ω
( )2ε ω
host
inclusions ( )mixε ω
( )mixD Eε ω=
Subject of intense investigations since mid-1800s: Mossotti, Clausius, Lorentz & Lorentz, Rayleigh, Garnett, Brugemann, Onsager, Wiener,… - see A. Sihvola, Electromagnetic mixing formulas and applications, 2008
Mixing dielectrics – “simple” way • Material density is computed as mass of mixture divided by volume (averaging) • May be simple permittivity averaging work for dielectrics?
38
mix i ii
vε ε= ∑ iv - volume fraction of material i
Works only for very limited number of cases! Example of failure in case of mixture with large difference of permittivities:
simple averaging ( )Re rε
~actual
tanδ simple averaging (pole at 15.8 GHz)
~actual (pole ~190 GHz)
1% of water in air; one-pole Debye model of water: 4.9; 76.1; 15.8rf GHzε ε∞ = ∆ = =
Frequency, Hz Frequency, Hz
Mixing dielectrics – right way • Average electric flux density and electric field!
39
mixD Eε=
2 2 1 1(1 )D v E v Eε ε= + −
1ε
2ε
2 1(1 )E v E v E= ⋅ + − ⋅
Example of fields averaging for spherical inclusions:
1E electric field in host
2E electric field in inclusions v volume fraction of inclusions 1
2 12 1
32
E Eεε ε
= ⋅+
Electric field in sphere:
( )2 1
1 12 1 2 1
32mix v
vε εε ε ε
ε ε ε ε−
= ++ − −
Maxwell Garnett mixing formula (derived by James Clerk Maxwell Garnett, 1880-1958):
J.C.Maxwell Garnett, Colours in metal glasses and metal films, Trans. of the Royal Soc., CCIII, 1904, p. 385-340.
1E2E
Field distortions is zero on average!
1
V
F F dvV
= ⋅∫
Bounds on permittivity of mixtures • Bounds for statistically homogeneous and isotropic mixture
40
1ε
2ε
volume fraction of inclusions
Z. Hashin, S. Shtrikman, “A variational approach to the theory of the effective magnetic permeability of multiphase materials,” J. Appl. Phys., vol. 33, no. 10, pp. 3125–3131, 1962.
,max 2
1 2 2
11
3
effv
vε ε
ε ε ε
−= +
+− ⋅
,min 1
2 1 1
1 13
effv
vε ε
ε ε ε
= +−+
− ⋅
1 2ε ε<
v
( )2 1
1 12 1 2 1
32mix v
vε εε ε ε
ε ε ε ε−
= ++ − −
Hashin-Shtrikman bounds are based on a variational treatment of the energy functional (3D):
These are the Maxwell Garnett’s equations!
Example (glass in resin): 1 23; 5;ε ε= =
volume fraction
,maxeffε
,mineffε
Very close!
Bounds on permittivity of mixtures • The loosest bounds for isotropic mixture defined by Otto Wiener
41
1ε
2ε
volume fraction of inclusions
O. Wiener, “Zur theorie der refraktionskonstanten,” Berichteüber Verhandlungen Königlich-Sächsischen Gesellschaft Wisseschaften Leipzig, pp. 256–277, 1910.
v
Wiener bounds are calculated for structured cases
Example (glass in resin):
volume fraction
,maxeffε
,mineffε
( )1 2
,min1 21eff v vε εε
ε ε⋅
=⋅ + − ⋅
( ),max 2 11eff v vε ε ε= ⋅ + − ⋅
Considerable difference!
1ε
2ε
1 23; 5;ε ε= =
parallel rule
series rule
Mixing with dispersion • Wiener bounds for mixture with 2 components
42
Example (glass in resin): Debye models 1 23; 0.2; 1 ; 5; 0.1; 100r rf GHz f GHzε ε ε ε∞ ∞= ∆ = = = ∆ = =
0.25v = 0.5v = 0.75v =
max min max min
max min
Actual model is somewhere between the bounds – may be considerable difference!
ε∞ (0)ε
Homogenization scale – feature size
43
• Homogenization area must be much smaller than the analyzed feature size • Dielectric inhomogeneity in cross-section may cause signal degradation at higher data
rates or frequencies – skew, mode conversion, anisotropy…
Homogeneous effective dielectric
more glass fiber more resin
effε
Layered effective dielectrics
More and more details is required to extend model frequency range…
1effε
2effε
1effε
Imbalanced effective dielectrics
2effε
Example of worst case analysis
44
model for resin model for glass Mixture model with 28% average volume content of glass in resin
~22.4% of glass ~33.6% of glass
8.5 mil 10.5 mil
more resin more glass
more resin
more glass
skew: ~ 3.5 ps/inch
Substantial far end common to differential mode transformation
See more at: Y. Shlepnev, C. Nwachukwu, “Modelling jitter induced by fibre weave effect in PCB dielectrics”, Proc. of 2014 IEEE Int. Symp. on EMC, 2014.
Model with +- 20% imbalance of glass in resin
Computed with Simbeor THz
Homogenization scale – wavelength
45
• Homogenization area must be much smaller than the wavelength • Effect of inhomogeneity along traces grow with frequency – skew, resonances…
more glass fiber at humps more resin in valleys
1D or 2D non-uniform t-line models
3D models
2116 20 mil
Wavelength in dielectric: 1 GHz – 6 in; 10 GHz – 600 mil; 50 GHz – 120 mil; 100 GHz – 60 mil;
Resonance at Period = Wavelength/2
Periodic change of dielectric properties
Example of periodicity effect analysis
46
model for resin model for glass
See more at: Y. Shlepnev, C. Nwachukwu, “Modelling jitter induced by fibre weave effect in PCB dielectrics”, Proc. of 2014 IEEE Int. Symp. on EMC, 2014.
~33.6% of glass
~22.4% of glass
Mixture model with 28% average volume content of glass in resin
Model with +- 20% sinusoidal periodic imbalance of glass in resin
Microstrip structure: traces at 9 deg.; Period 120 mil, resonance at ~32 GHz Reflective resonance -
No absorption as in Lorentzian dielectric model!
Reflection
Transmission
Computed with Simbeor THz
Anisotropic dielectrics
47
xx xy xz
yx yy yz
zx zy zz
ε ε εε ε ε ε
ε ε ε
=
xx xy xzt
xy yy yz
xz yz zz
ε ε εε ε ε ε ε
ε ε ε
= =
0 00 00 0
xx
yy
zz
εε
ε
( )0 0 1D E P Eε ε χ= ⋅ + = + ⋅
• Anisotropy is dependency of polarization on electric field direction
“Anisotropic solid is not an isotropic solid” – Lord Kelvin, 1904
0 00 00 0
εε
ε
=
=
⊥
D Eε= ⋅0Eε
P
x y
z Permittivity is 3x3 matrix, dyadic or second-rank tensor – 9 dispersive parameters in general
D
1. Reciprocal material – 6 parameters or less
Practically all anisotropic dielectrics are reciprocal
2. Biaxial material – 3 parameters
3. Uniaxial material – 2 parameters
Orthorhombic (monoclinic, triclinic) lattices and PCB laminates!
Tetragonal, hexagonal, rhombohedral lattices and PCB laminates!
x and y are identical, z is different
T.G. Mackay, Electromagnetic Anisotropy and Bianisotropy: A Field Guide, 2006
Anisotropy: biaxial dielectric • Homogenization of PCB dielectric along the coordinate axes
48
Orthorhombic system with optical axes as coordinate axes:
x y
z a b c
a != b !=c
z zz zD Eε= ⋅Series mixing rule (Weiner min)
Fiber glass fabric with different filling and warp yarns:
V
V V
y yy yD Eε= ⋅ x xx xD Eε= ⋅
x y
Anisotropy: uniaxial dielectric • In and out of plane homogenization of PCB dielectric
49
Tetragonal system with optical axes as coordinate axes:
x y
z a
a c
a != c
Series mixing rule (Weiner Min)
Fiber glass fabric with similar filling and warp yarns:
V
V
, ,x y x yD Eε== ⋅
z zD Eε⊥= ⋅
Out of plane value:
In plane value:
This value is usually is measured with wide strip line resonator (in spreadsheets)!
Parallel mixing rule (Weiner Max)
( )1 2
,min1 21eff v vε εε
ε ε⋅
=⋅ + − ⋅
( ),max 2 11eff v vε ε ε= ⋅ + − ⋅
M.Y. Koledintseva, S. Hinaga, and J.L. Drewniak, “Effect of anisotropy on extracted dielectric properties of PCB laminate dielectrics”, IEEE Symp. on EMC, Long Beach, CA, Aug. 14-19, 2011, pp. 514-517
Fields in PCB structures • X, Y and Z components of electric field depend on geometry
50
E-field in wide strip (25 Ohm at 10 GHz)
E-field in narrow strip (50 Ohm at 10 GHz)
E-field of strip differentia mode (85 Ohm at 10 GHz)
E-field in differential vias
Effective permittivity will depend on geometry too (it is averaging of the fields)
Computed with Simbeor THz
Alternative to anisotropic model • Layered dielectric model
51
Polarization current between 2 diff. vias at 10 GHz
Strip model with “resin” layer
Via model with “resin” and “glass” mixture layers
Substantial difference in current through layers with different permittivity
Computed with Simbeor THz
Which model is better for PCB?
52
Homogeneous Anisotropic Layered
Simplest – one permittivity 2 permittivities 2 or more permittivities and layer thicknesses
Depends on geometry, multiple models may be required for different cross-sections, vias,…
Accurate for extended range of geometries
Less accurate if feature size is smaller than the homogenization area (close traces or elements of vias)
Most accurate and universal
mixD Eε= 0 00 00 0
x x
y y
z z
D ED ED E
εε
ε
=
=
⊥
=
If dielectric components have substantially different permittivities:
Conductor dispersion effects • Current crowding below strips
• Around 10-100 KHz • Increases R and decreases L at very low frequencies
• Skin-effect • Transition frequencies from 1 MHz to 100 GHz (see chart) • Surface impedance boundary conditions (SIBC) for well-
developed skin-effect – R and L ~ sqrt(frequency)
• Skin-effect on rough surface • May be comparable with skin depth starting from 10 MHz • Increases both R and L (and possibly C)
• Ferromagnetic resonances (Nickel) • Plasmonic effects above 1 THz – (Drude model)
( ) (1 )2
Z iω µωσ⋅
= +
freq
uenc
y
( ) ( )rK Zω ω
R i Lω+
53
Skin effect = Maxwell’s eq. +Ohm’s law
BEt
∂∇× = −
∂
DH Jt
∂∇× = +
∂
J Eσ=
y yJ Eσ=Hz
Poynting’s vector
( )1expy s
s
iE E x
δ− +
= ⋅
x 1
s fδ
π µσ= Skin depth
Plane-wave view:
Current cancelation:
54
Example: currents in microstrip 1 KHz - skin depth 82*t 1 MHz - skin depth 2.6*t 10 MHz - skin depth 0.82*t
100 MHz - skin depth 0.26*t
t=1 mil, w=7 mil, current density in [A/m^2], 1V + 50 Ohm excitation
1 GHz - skin depth 0.082*t 100 GHz - skin depth 0.0082*t; peak current density in cross-section:
Computed with Simbeor THz 55
Current reversal in conductor
Currents at some distance from strip surface are flowing in opposite direction!
[A/m^2] 1 GHz - skin depth 0.082*t
56
Reversed current inside the strip Strip
TEM wave propagation direction (green arrows)
Delay of the wave propagating into the strip explain the current reverse and the internal inductance 1 GHz, Skin Depth 0.082*t (conductor thickness is 12.2 of SD)
[A/m^2]
Wave propagation direction inside conductor
Current reversal in conductor
57
Current reversal in conductor
Magnitude
Real
Imaginary
Strip Bottom Strip Top
Current Density at 1 GHz Real negative part means direction opposite to the surface currents! Similar to the current in round wire
Wire Center
Wire Radius
Magnitude Real
Imaginary
Reversed current
Reversed current
Skin depth
58
Skin-effect and roughness
40 MHz 150 MHz
4 GHz
0.5 um
400 GHz
Account for roughness
No roughness effect
10 um
5 um
1 um
18 GHz
0.1 um
Interconnect or plane thickness in micrometers vs. Frequency in GHz
RFIC
Package
IC
No skin-effect
Well-developed skin-effect
PCB
Roughness has to be accounted if rms value is comparable with the skin depth (0.5-1 of skin depth)
Transition from 0.5 skin depth to 2 and 5 skin depths for copper interconnects on PCB, Package, RFIC and IC
Ratio of skin depth to r.m.s. surface roughness in micrometers vs. frequency in GHz
59
Roughness modeling • Direct electromagnetic analysis is
simply not possible (very approximate) • “Effective dielectric roughness” layer • Roughness correction coefficients:
– Modified Hammerstad model – Huray’s snowball model – Hemispherical model – Sandstroem’s model – Stochastic models – Periodic frequency selective surfaces…
See references at: Y. Shlepnev, C. Nwachukwu, Practical methodology for analyzing the effect of conductor roughness on signal losses and dispersion in interconnects, DesignCon2012
Cross-section
Profilometer
60
Effective Roughness Dielectric (ERD)
Introduced in M.Y. Koledintseva, A. Ramzadze, A. Gafarov, S. De, S. Hinaga, J.L. Drewniak, PCB conductor surface roughness as a layer with effective material parameters. – in Proc. IEEE Symp. Electromagn. Compat., Pittsburg, PA, USA, 2012, p. 138-142.
1 1,eff tε
2 2,eff tε
copper t
Layer with mixture of conductor and dielectric material is turned into layer with “effective” dielectric parameters
Eliminates uncertainties of the conductor/dielectric boundary; Too many parameters, difficult to identify;
61
Example of analysis with ERD
With ERD Without With ERD
Without
ERD parameters for STD copper are defined in A.V. Rakov, S. De, M.Y. Koledintseva, S. Hinaga, J.L. Drewniak, R.J. Stanley, Quantification of conductor surface roughness profiles in printed circuit boards, IEEE Trans. on EMC, v. 57, N2, 2015, p. 264-273.
Causal increase in attenuation, phase delay and decrease in impedance!
ERD layer next to planes
ERD strips above and below copper strip Computed with
Simbeor THz
Attenuation IL
Phase Delay |Zo|
62
Modified Hammerstad model
( )21 arctan 1.4 1srs
K RFπ δ ∆
= + ⋅ ⋅ −
1s f
δπ µ σ
=⋅ ⋅ ⋅
Conductor skin-depth
RF - roughness factor, defines maximal growth of losses due to metal roughness (increase of surface)
∆ ~ root mean square peak-to-valley distance
Modified model suggested in Y. Shlepnev, C. Nwachukwu, Roughness characterization for interconnect analysis. - Proc. of the 2011 IEEE Int. Symp. on EMC, Long Beach, CA, USA, August, 2011, p. 518-523
E
HΠ Plane wave outside
“Absorption” waves on surface
1 mµ∆ =
RF=2 – original model
RF=3
RF=1.5
Frequency, Hz
∆
Roughness correction coefficient – increase of absorption by Ksr:
Bumps are much smaller than wavelength! 63
Huray’s snowball model
64
Losses estimation for conductive sphere are used to derive equation for multiple spheres:
122
22 21 1 s s
sr i ii i i
K A DD Dδ δ
−
= + ⋅ + +
∑
Amatte/Ahex can be accounted for by resistivity; Can be simplified to model with 2 parameters per ball (Ai and Di):
32
ii
hex
NAAπ
= Di – ball i diameter; Ni – number of balls with diameter Di;
P.G. Huray, The foundation of signal integrity, 2010
D= 2 um
D= 1.4 um
D= 1.7 um
A=2.1e12 1/um^2
Frequency, Hz
Dispersion with rough conductors
65
“Oliner’s waveguide – ideal to investigate RCCs
Copper: w=20 mil; t=1 mil; Rough; Ideal dielectric: Dk=4; h=5.3 mil;
PMC
MH: Del= 1 um; RF=2
Huray’s: A=2.1e12 1/um^2; D=1.7 um;
Attenuation, Np/m
Phase delay, s/m
Flat copper ~sqrt(f)
~f
Flat copper: Red lines; Huray’s one-ball: blue lines; Modified Hammerstad (MH): black lines;
Transition to skin-effect
Flat copper Frequency, Hz
Frequency, Hz
PMC
Use of roughness correction coefficients • Apply it to attenuation: Simplest; Non causal, applicable for t-lines only;
• Apply it to internal conductor part of p.u.l. impedance:
• Apply to conductor surface impedance operator (Simbeor)
66
( ) ( )r r sOhmZ f K Z i L
mω = ⋅ + ⋅ ∞
Kr is impedance roughness correction coefficient (Huray, Modified Hammerstad,…); Zs – conductor p.u.l. impedance matrix;
Simple, causal; Does not account for actual current distribution on conductor, applicable for t-lines only;
See details in Y. Shlepnev, C. Nwachukwu, Roughness characterization for interconnect analysis. - Proc. of the 2011 IEEE Int. Symp. on EMC, Long Beach, CA, USA, August, 2011, p. 518-523
" 1/2 1/2cs sr cs srZ K Z K= ⋅ ⋅
Ksr – diagonal matrix with roughness correction coefficients on diagonal (Huray, Modified Hammerstad,…); Zcs – conductor surface impedance operator (matrix);
Causal, accounts for actual current distribution; Difficult to implement, no capacitive effect; Boundary uncertainty in all approaches with RCC;
What is bulk resistivity?
Ferromagnetics: Nickel magnetization • Magnetic permeability dispersion equations are derived by
Landau and Lifshits from description moving boundaries of oppositely magnetized layers in ferromagnetic metal:
• Lorentz model may be also acceptable for resonance description • Can be combined with Debye model at lower frequencies and
Lorentz model at the millimeter frequencies L. Landau, E. Lifshits, On the theory of the dispersion of magnetic permeability in ferromagnetic bodies, Phys. Zeitsch. der Sow., v. 8, p. 153-169, 1935. Y. Shlepnev, S. McMorrow, Nickel characterization for interconnect analysis. - Proc. of the 2011 IEEE International Symposium on EMC, 2011, p. 524-529.
( ) ( )2
02 2
0 2h l hf i ff
f i f fγµ µ µ µ
γ+ ⋅ ⋅
= + − ⋅+ ⋅ ⋅ −
0
; ;[ ]; [ ]
l hpermeability at low frequencies permeability at high frequenciesf resonance frequency Hz damping coefficient Hzµ µ
γ− −− −
Frequency, Hz
67
68
Example: 150 mm microstrip link with ENIG finish with about 0.05 um of Au and about 6 um of Ni over the copper; Simulation with identified dielectric model and Landau-Lifshits model for Ni layer:
Simulated (green)
Measured (red)
Insertion Loss
Simulated (green)
Measured (red)
12 Gb/s Measured 12 Gb/s
Simulated
Breaking the skin: Drude model
69
freeJ Eσ= Bulk conductivity with temporal dispersion: ( ) 0
1 r
fi f fσσ =
+Relaxation frequency for copper is about ~18 THz, relaxation time ~9 fs
Skin depth Bulk conductivity
( )Im σ−
( )Re σ
With dispersion
Without
Good introduction: C.T.A. Johnk, Engineering electromagnetic – fields and waves, 1975
Frequency, Hz Frequency, Hz
• How to identify broadband dielectric model? • How to identify conductor roughness parameters? • How to separate dielectric, conductor and conductor roughness
models? • Can roughness losses be accounted in dielectric model? • Which roughness model is more accurate? • Other questions?...
Outstanding questions
70
Find some answers are in Simberian app notes at www.simberian.com
PCB materials and model identification techniques
• Composition of PCB Dielectric Materials • Overview of the material property identification
techniques • Identification with GMS-parameters
71
Presented by Chudy Nwachukwu, Isola
Composition of PCB Dielectrics
Just to name a few… Flex Polyimide
Flex Fluoropolymer / Polyimide composite
Liquid Crystal Polymer (LCP)
Ceramic Filled Polymer on Fiberglass
Glass Microfiber Reinforced PTFE
Micro-dispersed Ceramic in PTFE composite w/fiberglass
Ceramic filled PTFE on woven fiberglass
PTFE on woven fiberglass
Ceramic-filled Epoxy on fiberglass
High Tg Thermoset resin w/fiberglass reinforcement
Cross-section Images
72
Resin Chemistry – What’s in it?
Flame Retardants Brominated – Tetrabromobisphenol A (TBBA)
Low Halogen / Halogen Free
Phosphorous and Nitrogen based
Aluminum and Magnesium hydroxide
Filler components Aluminum Silicate
Talc
Rubber
Glass microspheres
Boron Nitride
Br Br
Br
Br
OH O
ROH
OO
CH3
CH3
R
O
73
Viscosity Regulator
High Shear Milling/Mixer
To the Treater
Filtration
Compounding / Mixing Process
Woven Filter – removes contaminants in liquid
components.
Magnetic Filter – removes ferrous contaminates.
High Shear Milling/Mixing – ensures homogenous mixing
of all components (solvent, catalysts, hardeners).
Viscosity measurement and feedback
Components
74
Composition – Fiberglass Weave
Property Low DK Low CTE Improves Degrades E-Glass D-Glass L-Glass NE-Glass T-Glass S-Glass
SiO2 DK / DF Drillability 52 - 56% 72 - 76% 52 - 56% 52 - 56% 64 - 66% 64 - 66% CaO DK 20 - 25% 0% 0 - 10% 0% 0% 0 - 0.3% Al2O3 DF 12 - 16% 0 - 5% 10 - 15% 10 - 18% 24 - 26% 24 - 26% B2O3 DK / DF 5 - 10% 20 - 25% 15 - 20% 18 - 25% 0% 0% MgO Meltability DK 0 - 5% 0% 0 - 5% 5 - 12% 9 - 11% 9 - 11%
Na2O / K2O DK / DF / Drillability 0 - 1% 3 - 5% 0 - 1% 0 - 1% 0% 0 - 0.3%
TiO2 / LiO2 Meltability 0% 0% 0 - 5% 0% 0% 0%
Property Unit E-Glass Low DK Glass Low CTE Glass
DK Freq (1 GHz) 6.8 4.8 5.4 DF Freq (1 GHz) 0.0035 0.0015 0.0043
Tensile Modulus Gpa 75 64 86
Thermal Expansion ppm/⁰C 5.6 3.3 2.8
75
Fabric Manufacturing Process
Quality Inspection
Hollow Fibers
Yarn twist
Broken Filaments
Impurities
Critical Measures
Woven glass styles
Electrical properties
PCB Process-ability
Cost & Availability
76
B-Stage Treating
Glass Cloth
Unwinders
Splicer Pre-dip & Primary Dip Pans
Ovens Rewinder Sheeter Stacker
Prepreg in roll form
Prepreg in sheets/panels
Inspection
Accumulator & Tension Control
77
Material Identification Techniques • For test structures …
– Sample in transmission or resonant structure – Transmission line segment or resonator made with the material
• Make measurements … – Capacitance – S-parameters measured with VNA – TDR/TDT measurements – Combination of measurements
• Correlated with a numerical model – Analytical or closed-form – Static or quasi-static field solvers – 3D full-wave solvers
78
Characterizing “Effective” Permittivity
Copper-clad Dielectric Testing Short Pulse Propagation (SPP)
Generalized Modal S-Parameter
(GMSP)
Unclad Dielectric Testing Capacitance Test Method
Coupled Stripline “Berezkin”
Resonant Cavity Structures
Free-space Transmission
79
Capacitance Test Method (1 MHz – 1 GHz)
Parallel Plate Fixture Admittance is modeled as parallel “G” || “C”
Capacitance is modeled as parallel plate “C”
Effect of fringing fields are neglected.
Presence of dielectric sample changes
impedance of the parallel plate capacitor.
Accuracy for the test method is critically
dependent on thickness uniformity of the
dielectric sample.
80
Coupled Stripline Fixture (1 GHz – 22 GHz)
81
Resonant Cavity Methods (3 GHz - 40 GHz)
Split Post Cavity Each cavity is designed with a specific Q factor and
measures in-plane dielectric permittivity.
Discrete frequency measurements (example: 3, 7, 10,
15.5 & 22.5 GHz).
Open Resonator
Courtesy of Damaskos Inc. 82
Free-space Quasi Optical (18 GHz – 110 GHz)
Measurement Steps: Isolation – blocking the beam propagation path
with a metal plate to account for diffraction
effects residual reflections.
Reference – measuring through transmission
(S21) parameters without material under test to
account for the permittivity contributions of air.
Time domain gating – Mathematical elimination
of multipath signals using the sum of distance
between horn antennas and dielectric sample
(eg: +/- 2ns).
83
Sample data from Unclad Dielectric testing
84
Short Pulse Propagation (SPP)
Low frequency values are identified separately due to TDT limitations
TDT pulse responses of 2 line segments -> Gamma (complex propagation constant)
Iterative matching of measured and computed Gamma -> Dielectric Model
A. Deutsch, T.-M. Winkel, G. V. Kopcsay, C. W. Surovic, B. J. Rubin, G. A. Katopis, B. J. Chamberlin, R. S. Krabbenhoft, Extraction of and for printed circuit board insulators up to 30 GHz using the short-pulse propagation technique, IEEE Trans. on Adv. Packaging, vol. 28, 2005, N 1, p. 4-12.
85
GMS-Parameters
See details at: Y. Shlepnev, A. Neves, T. Dagostino, S. McMorrow, Practical identification of dispersive dielectric models with generalized modal S-parameters for analysis of interconnects in 6-100 Gb/s applications, DesignCon 2009, available at www.simberian.com Y. Shlepnev, PCB and package design up to 50 GHz: Identifying dielectric and conductor roughness models, The PCB Design Magazine, February 2014, p. 12-28.
L
Optimization loop – red line; Automated in Simbeor software;
( )( )
0 expexp 0
LGMSc L−Γ ⋅ = −Γ ⋅
86
Comparison of GMS and SPP techniques
87
• Commonalities: – Same test fixture can be used (2 segments) – Numerical transmission line model is used in both techniques – Resistance measurement at DC can be used to identify bulk resistivity in both techniques
• Differences: – Measured S-parameters are used to extract GMS-parameters (VNA), but short pulse TDT
measurements are used in SPP technique to extract complex propagation constants – SPP uses measurements at 1 MHz to have low frequency asymptotes of dielectric constant - not
needed with the GMS-parameters if S-parameters are measured starting from sufficiently low frequency
• If S-parameters are used to extract Gamma from GMS-parameters, such technique may be considered as a variation of SPP methodology – “SPP Light”
– Identification with GMS-parameters and “SPP Light” should produce nearly identical results if same t-line model is used
Details in Y. Shlepnev, Broadband material model identification with GMS-parameters, EPEPS 2015.
Example of identification
88
From Isola FR408HR specifications
10.5 (11) mil strip lines; microstrips 13.5 (14.5) mil; Use measured S-parameters for 2 segments (2 inch and 8 inch); No data for conductor roughness model;
CMP-28 channel modelling platform from Wild River Technology http://www.wildrivertech.com/
Identification with GMS and SPP
89
• Dielectric: Wideband Debye dielectric model with Dk=3.8 (3.66), LT=0.0117 @ 1 GHz;
• Conductor roughness: modified Hammerstad model with SR=0.32 um, RF=3.3
GMS-parameters Gamma (SPP Light)
Models are usable above 50 GHz!
Models identified with GMS-parameters
Data from W. Beyene et al., Lessons learned: How to make predictable PCB interconnects for data rates of 50 Gbps and beyond, DesignCon 2014. 90
Implication of Material Characterization Methods
91
Practical PCB Material Identification Techniques
92
Presented by Scott McMorrow, Samtec-Teraspeed
Wideband Debye model properties
4.2rε =
Frequency, Hz
( )Re rε
( )tanδ ω
POLES 110m 210m
Dk and LT at one point is sufficient to define the model!
tan 0.02δ =90 10f Hz=
93
Djordjevic-Sarkar model assumptions • Dielectric properties represent the behavior
of two poles • Low frequency pole (kHz) • High frequency pole (THz) • Well outside the frequency band
that we want to characterize for data transmission.
Djordjevic-Sarkar model advantages • Describes most materials used in
PCB/Package/Cable • Simple to adjust
1 MHz 100 GHz
Plane wave in Wideband Debye dielectric
Attenuation Np/m
Phase delay, s/m
Frequency, Hz 94
~f
1 MHz 100 GHz
Both attenuation and phase delay provide the same information regarding the dielectric loss. Slope of the phase delay is dependent upon loss tangent. We can use this to identify dielectric, since there is a fairly sensitive slope.
Practical implication of rough conductors
95
“Oliner’s waveguide – ideal to investigate RCCs
Copper: w=20 mil; t=1 mil; Rough; Ideal dielectric: Dk=4; h=5.3 mil;
PMC
Attenuation, Np/m
Phase delay, s/m
Flat copper ~sqrt(f)
~f
Transition to skin-effect
Flat copper
Frequency, Hz
PMC
1 MHz 100 GHz
Roughness has a large impact on loss. Roughness has a very small impact on phase delay. We can use this in the final tuning of overall interconnect loss. We can neglect roughness for the purpose of identifying Dk and Df.
GMS-Parameters
See details at: Y. Shlepnev, A. Neves, T. Dagostino, S. McMorrow, Practical identification of dispersive dielectric models with generalized modal S-parameters for analysis of interconnects in 6-100 Gb/s applications, DesignCon 2009, available at www.simberian.com Y. Shlepnev, PCB and package design up to 50 GHz: Identifying dielectric and conductor roughness models, The PCB Design Magazine, February 2014, p. 12-28.
L
Optimization loop – red line; Automated in Simbeor software;
( )( )
0 expexp 0
LGMSc L−Γ ⋅ = −Γ ⋅
96
Raw vs. GMS
97
Common Mode
Differential Mode Noise occurs when return loss crosses insertion loss
Differences between uniform sections of two measurements become apparent with GMS
technique
Filtered vs. Unfiltered Attenuation
Common Mode
Differential Mode
Mode separation occurs when dielectric is not uniform when working with differential conductors. The position of Differential Mode vs Common Mode provides information on where the difference
occur
Unfiltered Phase Delay
Common Mode
Differential Mode
Mode separation occurs when dielectric is not uniform when working with differential conductors. The position of Differential Mode vs Common Mode provides information on where the difference occurs. Faster Common Mode indicates common mode fields are exposed to a lower Dk dielectric.
Phase Delay is always much cleaner than
Attenuation or Group Delay
Comparison of GMS and AFR
GMS-parameter method is designed to remove losses due to impedance mismatch by normalizing to a perfectly matched condition at every frequency point.
Other methods are designed to create faithful models of the actual delta-length interconnect. This may introduce additional losses as mismatch increases.
Red - Differential Mode GMS Dark Green – Differential Mode AFR Light Green – Common Mode AFR
Blue – Common Mode GFS
Comparison of GMS and AFR Phase Delay
Common Mode
Differential Mode
Essentially identical delay between GMS and AFR methods.
Phase or Phase delay is generally the most stable method for identifying dielectric properties.
Modeled vs. Measured Phase Delay
Modeled vs. Measured Attenuation
Trace Geometry Cross Section
Resin rich / Fiber Free Region
Differential Pair Geometry
Resin rich / Fiber Free Region
To correctly model differential trace geometries, anisotropic layering must be modeled. Resin/Epoxy/Polymer regions are always lower Dk than mixed dielectric regions. Laminate weave skew is identified and bounded
through measurements and then incorporated into channel models as a post process step.
Dielectric Mixture Modeling
Dielectric average of epoxy and glass
Pure epoxy
Difference between epoxy Er and Average Er results in separation of common and
differential propagation modes.
Measured Meg6 Diff Stripline
Insertion Loss / Return Loss crossover @ 13 GHz
Forward Crosstalk
Measured data is often limited by Signal-to-Noise ratio at the insertion loss / return loss crossover point. But even this data can produce good model correlation if parameters are extracted between DC and 13 GHz.
Meg 6 Mode Separation Phase
Differential Mode is Faster
Common Mode is Slower
Mode separation due to layered anisotropy of epoxy and fiber rich areas in laminate system
Meg 6 Mode Separation Group Delay
Differential Mode is Faster
Common Mode is Slower
Teraspeed Consulting Group LLC
Mode separation due to layered anisotropy of epoy and fiber rich areas in laminate system.
Megtron 6 20” Differential Pair Modeled vs. Measured Single-ended S-parameters
Practical Material Identification
• Step 1 – Use group/phase delay for preliminary Er • Step 2 – Evaluate potential variation • Step 3 – Identify low frequency characteristics • Step 4 – Adjust for dielectric loss • Step 5 – Final adjustment for conductor roughness
• 111
Practical Material Identification Step 1 – Group Delay Preliminary Er
Identification
112
Tune Dk near 1 GHz to match phase
Practical Material Identification Step 2 – Evaluate variation
113
Yuck!
Practical Material Identification Step 3 – Identify Low Frequency Characteristics
114
Adjust for conductivity
Practical Material Identification Step 4 – Adjustment for Dielectric Loss
115
Tune Df to match phase at high frequency
Practical Material Identification Step 5 – Final Adjustment for Conductor Roughness
116
Tune roughness model to match high frequency loss.
Terragreen Raw Measurements
Terragreen Phase Delay GMS vs Modeled
Ansys SiWave and Simbeor have very faithful match to GMS-parameter
plot
Terragreen Attenuation GMS vs Modeled
Ansys SiWave and Simbeor have very faithful match to GMS-parameter
plot Dk = 3.587 Df = 0.004
Roughness = 0.285 um Roughness Factor = 2
Tachyon 100G Measured Insertion Loss
2 inch Horiz / Vert 0 degree
4 inch Horiz / Vert 4.5 degree
6 inch Horiz /Vert 0 degree
8 inch Horiz / Vert 4.5 degree
8 inch Horiz 4.5 degree Horizontal Weave Periodic Loading
Variation of Dk in horizontal weave direction is discerned by 4.5 degree periodic weave loading, which causes a ½ wave resonance at ½ the
crossing frequency
Tachyon 100G 4” Generalized De-embedded Attenuation Match
Red – Vertical (Fill) Direction Blue – Horizontal (Warp) Direction
Black – Simulated Attenuation
8 inch Horiz 4.5 degree Horizontal Weave Periodic Loading
Cu Conductivity – 5.6 e7 S/M Cu Roughness – 0.4 micron (Hamerstadt-Jensen)
Dk – 3.06 @ 1 GHz (Djordjevic-Sarkar) Df - .0025 @ 1 Ghz (Djordjevic-Sarkar)
Due to large difference between Dk of polymer and glass, Tachyon is extremely sensitive to dielectric
variations in all directions
Material Comparison De-embedded Periodic Weave Resonance
Light Blue – Tachyon Pink – Terragreen
Green – I-Tera Blue – Megtron 6 Purple – I-Speed Red – Megtron 4
Orange – Gigasync RTF Brown – Gigasync H-VLP
Gigasync exhibits a breakpoint in loss characteristics around 7-10 GHz, where the loss slope changes for both
RTF and H-VLP copper. Indicatesan additional pole in the material dielectric
response that is not predicted by the Djordjevic-Sarkar model.
Periodic weave resonance is only discerned in Horizontal (Warp) direction. Vertical (Fill) direction shows no evidence of this phenomena.
Traces in Horizontal (Warp) direction will still experience variation in Dk
based upon local weave environment.
Modeled
Modeled with 12 ps launch skew added for laminate weave skew
Megtron 6 20” Differential Pair Modeled vs. Measured Differential S-parameters
Laminate weave skew introduces an additional factor in the assessment of models. In this case, one measured set of differential pairs had significant P/N skew of 12 ps, identified in group delay and phase plots.
QUESTIONS?
Thank you!
MORE INFORMATION:
www.isodesign.isola-group.com
Chudy.Nwachukwu@isola-group.com
www.simberian.com
Shlepnev@simberian.com
www.teraspeed.com
Scott@teraspeed.com
top related